Iterate efficiently over a list of strings to get matrix of pairwise WMD distances

Question

I am trying to generate a matrix of pairwise distances from a list strings (newspaper articles).

WMD distance is not implemented in scipy.spatial.distance.pdist so I hook this implementation: https://github.com/src-d/wmd-relax to SpaCy. However, I cannot figure out how to iterate over my list to generate the distance matrix.

Answer 1

As per doc:

import spacy
import wmd
import numpy as np


nlp = spacy.load('en_core_web_md')
nlp.add_pipe(wmd.WMD.SpacySimilarityHook(nlp), last=True)

# given articles is a list of strings
docs = [nlp(article) for article in articles]

# matrix is just a list of lists in terms of Python objects
m = []
for doc1 in docs:
    row = []
    for doc2 in docs:
        # if distance is similarity function
        row.append(doc1.similarity(doc2))
    m.append(row)

result = np.matrix(m)

Numpy matrix doc